VC-1 zero block detection method in motion estimation video coding

Abstract

In this paper, we propose a VC-1 zero block detection method, for very low bit-rate applications that skip the transform operation for detected zero blocks. The proposed method works during the process of motion search for inter frame coding, by comparing the SAD of each residual block to two thresholds based on a value related to the QP, resulting in a two-option method for each threshold; where the 1st is for preserving quality, and the 2nd is for saving time. Normally VC-1 skips the quantization operation for detected zero blocks. In addition when a MB (macro-block) is detected as a zero MB, meaning that all 6 blocks in a MB are zeros; VC-1 signals this MB as a skipped MB, where it's treated differently and many processes are skipped on the MB layer. For that reason, the proposed method is enhanced again with skipping the transform operation, but for MBs with five zero blocks. Meaning that a 6th non-zero block in the MB will be a zero block, and the whole MB will then be signaled as a skipped MB. Loosing the information of the non-zero block, adds quality degradation, but compression takes place more effectively. Among the four sequences tested the Hall showed the best results with up to 73.55%, 78.00%, and 88.91% of the original file size, encoding, and decoding times respectively. On the other hand, the quality degraded, as was expected, with a maximum average PSNR by 5.3 dB for the Hall sequence and 3.4 dB in the Foreman sequence.